A practical solution to pseudoreplication bias in single-cell studies

Abstract

Cells from the same individual share common genetic and environmental backgrounds and are not statistically independent; therefore, they are subsamples or pseudoreplicates. Thus, single-cell data have a hierarchical structure that many current single-cell methods do not address, leading to biased inference, highly inflated type 1 error rates, and reduced robustness and reproducibility. This includes methods that use a batch effect correction for individual as a means of accounting for within-sample correlation. Here, we document this dependence across a range of cell types and show that pseudo-bulk aggregation methods are conservative and underpowered relative to mixed models. To compute differential expression within a specific cell type across treatment groups, we propose applying generalized linear mixed models with a random effect for individual, to properly account for both zero inflation and the correlation structure among measures from cells within an individual. Finally, we provide power estimates across a range of experimental conditions to assist researchers in designing appropriately powered studies.

NEBULA is a fast negative binomial mixed model for differential or co-expression analysis of large-scale multi-subject single-cell data

“…One potential disadvantage of this strategy is that it cannot adjust for cell-level covariates such as mitochondrial and ribosomal gene concentration to increase the statistical power. The pseudo-bulk method might also be underpowered when the CPS values are imbalanced across the subjects 41 .…”

Diverse human astrocyte and microglial transcriptional responses to Alzheimer’s pathology

“…The statistical analysis was performed using the limma package in R 64 , using the default configuration and the following linear model: ~pathology+nFeature+pc_mito, where pathology is the average immunohistochemistry quantification value for Aβ or pTau, nFeatures is the total number of distinct features expressed in each nucleus (to account for the fact that nuclei that express a higher number of features may have higher AUCell scores) and pc_mito is the percentage of counts mapping to mitochondrial genes. We also corrected for a potential pseudoreplication bias 65 , by using the duplicateCorrelation function of the limma package with the sample as the "blocking" variable.…”

scFlow: A Scalable and Reproducible Analysis Pipeline for Single-Cell RNA Sequencing Data

“…Log2, CPM), and co-variate scaling and centering. The default DGE method in scFlow is a generalized linear mixed model (GLMM) with a random effect (RE) term (e.g., to account for correlations within individual samples) as implemented within the model-based analysis of single-cell transcriptomics (MAST) algorithm (Zimmerman et al, 2021;Finak et al, 2015). An interactive DGE HTML report with a volcano plot and searchable tables is generated, including details of model parameters, inputs, and outputs (File 5).…”

“…A log2(TPM + 1) expression matrix is calculated from the raw counts matrix, and a two-part (i.e., including a discrete logistic regression component for expression rate and a continuous Gaussian component conditioned on each cell expressing a gene) generalized regression model is fit independently for each gene. The CDR is included as a covariate alongside additional user-specified experimental covariates, which can include, for example, the individual sample as a random effect (Zimmerman et al, 2021). False-discovery rate (FDR) adjusted p-values are determined using the Benjamini & Hochberg method.…”